Research On Feature Selection Method Based On Multi-objective Optimization And Its Application In Hypertension Prediction

With the coming of the age of aging,the health problems of the elderly have attracted more and more attention.Among them,non-communicable diseases such as hypertension have become the primary cause of threatening the health of the elderly.In order to reduce the incidence of hypertension in the elderly population,the influencing factors of hypertension have been widely studied in recent years.There are various data related to hypertension in the fields of clinical medicine and public health.Most of the research on the factors affecting hypertension in the past is to use traditional statistical methods.However,with the arriving of the era of big data,traditional statistical method has many limitations in high-dimensional data modeling;at the same time,machine learning algorithms are beginning to emerge,and algorithms such as random forests are widely used in various fields,but machine learning also exists some difficults with high-dimensional data modeling.So,a large number of feature selection methods were born in these circumstances.Feature selection is often seen as an essential weapon for high-dimensional data.How to design better feature selection schemes to better handle high-dimensional data is worth exploring.In this context,this study proposes a combined feature selection method based on multi-objective optimization.Usually in the research of Combinatorial Feature Selection methods,the number of feature selection in the first stage is set artificially.This study attempts to introduce a trade-off index,considering the stability of feature selection method and the fitting performance of machine learning algorithm,to give the screening criteria for the number of feature selection in the first stage.In the second stage,multi-objective genetic algorithm uses the feature subset provided in the first stage as the initial population to further feature selection,and jointly optimize the number of feature subsets and fitting performance of machine learning algorithm.Based on the data of China Health and Old-age Follow-up Survey,we made an empirical analysis to verify the validity of the combined feature selection method based on multi-objective optimization and the rationality of introducing trade-offindicators,aiming at predicting the social influencing factors of hypertension.Four different feature selection methods were combined with multi-objective genetic algorithm to establish four combinatorial feature selection schemes based on multi-objective optimization.Experiments show that the combined feature selection method is more effective than the single feature selection method,that is,the prediction error is smaller and the number of feature subsets is smaller.The validity of the combined feature selection method based on multi-objective optimization and the introduction of trade-off index in the first stage are verified to be scientific and reasonable.We believe that the introduction of trade-off indicators in combinatorial feature selection can provide some exploratory ideas for future research in the field of Combinatorial Feature selection.In addition,this method can not only be used to predict the influencing factors of hypertension,but also make more attempts in other practical application fields.